Probing 3D and NLTE models using APOGEE observations of globular cluster stars - Instituto de ...

Page created by Paula Brown
 
CONTINUE READING
Astronomy & Astrophysics manuscript no. 39484corr                                                                                  ©ESO 2021
                                               January 29, 2021

                                                    Probing 3D and NLTE models using APOGEE observations of
                                                                      globular cluster stars
                                                  T. Masseron1, 2 , Y. Osorio1, 2 , D.A.García-Hernández1, 2 , C. Allende Prieto1, 2 , O. Zamora1, 2 , and Sz. Mészáros3, 4

                                                    1
                                                        Instituto de Astrofísica de Canarias, E-38205 La Laguna, Tenerife, Spain
                                                    2
                                                        Departamento de Astrofísica, Universidad de La Laguna, E-38206 La Laguna, Tenerife, Spain
                                                        e-mail: tmasseron@iac.es
                                                    3
                                                        ELTE Eötvös Loránd University, Gothard Astrophysical Observatory, 9700 Szombathely, Szent Imre H. st. 112, Hungary
                                                    4
                                                        MTA-ELTE Exoplanet Research Group
arXiv:2101.11643v1 [astro-ph.SR] 27 Jan 2021

                                                    5
                                                        Premium Postdoctoral Fellow of the Hungarian Academy of Sciences
                                                    Received ; accepted

                                                                                                                ABSTRACT

                                                    Context. Hydrodynamical (or 3D) and non-local thermodynamic equilibrium (NLTE) effects are known to affect abundance analyses.
                                                    However, there are very few observational abundance tests of 3D and NLTE models.
                                                    Aims. We developed a new way of testing the abundance predictions of 3D and NLTE models, taking advantage of large spectroscopic
                                                    survey data.
                                                    Methods. We use a line-by-line analysis of the Apache Point Observatory Galactic Evolution Experiment (APOGEE) spectra (H band)
                                                    with the Brussels Automatic Code for Characterizing High accUracy Spectra (BACCHUS). We compute line-by-line abundances of
                                                    Mg, Si, Ca, and Fe for a large number of globular cluster K giants in the APOGEE survey. We compare this line-by-line analysis
                                                    against NLTE and 3D predictions.
                                                    Results. While the 1D–NLTE models provide corrections in the right direction, there are quantitative discrepancies between different
                                                    models. We observe a better agreement with the data for the models including reliable collisional cross-sections. The agreement
                                                    between data and models is not always satisfactory when the 3D spectra are computed in LTE. However, we note that for a fair
                                                    comparison, 3D corrections should be computed with self-consistently derived stellar parameters, and not on 1D models with identical
                                                    stellar parameters. Finally, we focus on 3D and NLTE effects on Fe lines in the H band, where we observe a systematic difference in
                                                    abundance relative to the value from the optical. Our results suggest that the metallicities obtained from the H band are more accurate
                                                    in metal-poor giants.
                                                    Conclusions. Current 1D–NLTE models provide reliable abundance corrections, but only when the atom data and collisional cross-
                                                    sections are accurate and complete. Therefore, we call for more atomic data for NLTE calculations. In contrast, we show that 3D
                                                    corrections in LTE conditions are often not accurate enough, thus confirming that 3D abundance corrections are only valid when
                                                    NLTE is taken into account. Consequently, more extended self-consistent 3D–NLTE computations need to be made. The method we
                                                    have developed for testing 3D and NLTE models could be extended to other lines and elements, and is particularly suited for large
                                                    spectroscopic surveys.
                                                    Key words.

                                               1. Introduction                                                           of the lines. Its model atom is also fairly well established
                                                                                                                         now (Osorio et al. 2015) and also has good collisional cross-
                                               Accurate chemical analysis of cool stellar spectra (Teff .                sections. Therefore, there are a large number of NLTE computa-
                                               8000K) rely on a deep understanding of radiative transfer                 tions for this element in late-type stars (e.g. Merle et al. 2011;
                                               physics. As extensively explored over many astronomical ap-               Bergemann et al. 2015; Osorio & Barklem 2016; Zhang et al.
                                               plications in the review by Nissen & Gustafsson (2018), cur-              2017; Alexeeva et al. 2018). An important difference between
                                               rently the two main challenges for abundance derivation in-               the atom model Osorio et al. (2015) and other atom models pre-
                                               volve the three-dimensional (3D) modelling of hydrodynam-                 sented in later works lies in the adoption of collisional data in-
                                               ics of stellar atmospheres and the modelling of line formation            volving high-lying levels of Mg I (from which many IR lines
                                               in conditions of non-local thermodynamic equilibrium (NLTE)               are formed). Both Zhang et al. (2017) and Alexeeva et al. (2018)
                                               (see Nordlund & Stein 2009; Bergemann & Nordlander 2014;                  ignored hydrogen collisions involving high-lying levels of Mg
                                               Bergemann 2014, for the physics principles).                              I while Osorio et al. (2015) uses the formula from Kaulakys
                                               Among the large number of publications on NLTE effects (each              (1984). Electron collisional ionisation for high-lying levels of
                                               generally focused on one particular element and on a limited              Mg I were also updated in Osorio et al. (2015), where instead of
                                               number of lines; see review by Mashonkina 2014), some ele-                the classical formula (Seaton 1962) the method from Vrinceanu
                                               ments have been particularly studied.                                     (2005) was adopted.
                                                   Magnesium is the archetypal species for NLTE calcu-                       Calcium is fundamental for stellar spectroscopy, notably
                                               lations, notably because the Mg triplet in the solar opti-                because of the triplet in the near-IR that represents an in-
                                               cal spectrum clearly shows the NLTE signature in the core                 evitable feature for metallicity estimate and more particularly
                                                                                                                                                                    Article number, page 1 of 13
A&A proofs: manuscript no. 39484corr

for the Gaia mission. Therefore, a large number of NLTE stud-         Mashonkina et al. 2011; Lind et al. 2017). Another alternative
ies have been dedicated to this element (e.g. Merle et al. 2011;      consists in comparing the abundances obtained between neutral
Mashonkina et al. 2017). It now also has a quite accurate model       and ionised lines of the same element (e.g. Korn et al. 2003). On
atom and good collisional data (Osorio et al. 2019). Similar im-      the contrary, it is not so straightforward to test 3D effects against
provements to the Mg atom are present in the Osorio et al.            line profiles. Only the solar spectrum (Asplund et al. 2000a,b;
(2019) model atom.                                                    Caffau et al. 2008; Pereira et al. 2009; Lind et al. 2017) and Pro-
    Like Mg and Ca, silicon is an α-element, and thus its mea-        cyon’s spectrum (Allende Prieto et al. 2002) are used for direct
surement reflects the chemical signature of core-collapse super-      comparison against 3D line profiles.
novae nucleosynthesis. Estimates for NLTE of Si have been pub-        In summary, both 3D and NLTE line formation predictions have
lished for various stellar spectra (e.g. Bergemann et al. 2013;       been mostly tested only against a few available extremely high-
Mashonkina et al. 2016a; Zhang et al. 2016).                          resolution spectra with high signal-to-noise ratios. Moreover,
    Iron is fundamental in stellar spectroscopy; it is used           because absolute elemental abundances need to be fixed in the
as the main metallicity indicator due to the numerous lines           comparison, the solar spectrum turned out to be nearly the only
present in the spectra, which are also used to infer the main         reliable data to be used for testing.
stellar parameters (Teff and log g). Therefore, its NLTE effects          Meanwhile, several large spectroscopic surveys now offer
have been extensively studied in literature (Bergemann et al.         a large amount of stellar spectra (several hundred thousand)
2012a; Mashonkina et al. 2011, 2016b). However, despite its ex-       (APOGEE, Majewski et al. 2017; Gaia-ESO, Gilmore et al.
tremely deep implication in all stellar studies, its model atom       2012; Randich et al. 2013; GALAH, De Silva et al. 2015). These
and collisional cross-sections are currently both theoretically       surveys have enough resolution power to resolve individual
and experimentally incomplete, and sometimes empirical esti-          lines, and are therefore strongly affected by systematic errors
mates are necessary (Ezzeddine et al. 2018; Caffau et al. 2019;       due to NLTE and 3D effects that impact their 1D–LTE abun-
Korotin et al. 2020).                                                 dance determination. For example, a systematic discrepancy
    Despite the large amount of atomic data simultaneously re-        in the derived metallicity of globular clusters between the IR
quired, it is now possible to perform NLTE synthesis on a large       APOGEE data and optical literature has been found across the
scale (e.g. Kovalev et al. 2019; Osorio et al. 2020). However, the    various releases (Mészáros et al. 2013; Holtzman et al. 2015),
main limitations and uncertainties regarding NLTE calculations        and has been confirmed by independent analysis of the same
depend on the fundamental knowledge of the collisional data and       data (Mészáros et al. 2015; Masseron et al. 2019). This suggests
the structure of the atom.                                            that a 3D or NLTE effect can impact the Fe lines. Moreover,
In contrast, hydrodynamical simulations of stellar atmospheres        Hawkins et al. (2016) suspected that NLTE effects on some of
are still very expensive with regard to computational resources,      the Ti and Al lines are affecting the abundance measurements in
which means that currently the largest grids (Ludwig et al. 2009;     the APOGEE spectra. Unfortunately, despite these possible sig-
Magic et al. 2013) provided by the three main codes (MuRAM            natures of NLTE and/or 3D effects, the resolutions of the surveys
Vögler et al. 2005, CO5 BOLD Freytag et al. 2012, and Stagger         spectrographs do not usually allow a detailed line profile exami-
Galsgaard & Nordlund 1996) only include a few tens of models.         nation (see Fig. 1). Therefore, abundance results from these sur-
From these limited number of models, 3D corrections on indi-          veys more generally rely on 3D and NLTE models to correct the
vidual elements are still being estimated (e.g. Spite et al. 2012     abundances (e.g. Kovalev et al. 2019), and do not question them.
for Ca, Černiauskas et al. 2018 for Mg, and Collet et al. 2006           Still, despite the relatively limited resolution power of large
for several elements in a single star).                               spectroscopic surveys, we propose here an empirical method to
    Naturally, the ultimate goal regarding abundance determina-       test 3D and NLTE effects based on a line-by-line differential ap-
tion is to achieve consistent 3D and NLTE calculations. The first     proach. Although a few other studies have attempted to compare
developments have recently been made, but fully consistent 3D         line abundances to detect lines biased by substantial NLTE or 3D
and NLTE analysis1 are limited to a few elements in one star or       effects, they relied on a small number of spectra (e.g. ≈140 in
one element in a few stars (Asplund et al. 2003; Sbordone et al.      Roederer et al. 2014). Here we take advantage of the large num-
2010; Lind et al. 2013; Steffen et al. 2015; Klevas et al. 2016;      ber of observations of the APOGEE survey of globular clusters
Amarsi et al. 2016; Nordlander, T. et al. 2017; Amarsi et al.         (GCs) stars to quantify and discuss the occurrence of 3D and
2019). Those works illustrate that 3D and NLTE computed si-           NLTE effects in stellar spectra of cool metal-poor stars.
multaneously in a self-consistent manner clearly provides results
that are different to the simple sum of independently computed        2. Methodology
3D and NLTE effects. A more computationally efficient alterna-
tive consists in using a horizontal average of the 3D atmosphere      All abundance results in this work were determined in a homoge-
and then computing the NLTE corrections (Bergemann et al.             neous way using the Brussels Automatic Code for Characteriz-
2012b; Mashonkina et al. 2013), but this is still a limited approx-   ing High accUracy Spectra (BACCHUS) (Masseron et al. 2016).
imation (e.g. Lind et al. 2017).                                      In this version of the code, a pre-computed grid of 1D–LTE spec-
    This illustrates that different approaches exist for comput-      tra were compared to an observed spectrum spectrum (Sec. 2.1),
ing 3D and NLTE effects on line formation. There are vari-            a 3D spectrum and a NLTE spectrum (Sec. 2.2). While all other
ous prescriptions between the available models and codes, as          parameters were fixed, each of the 1D–LTE spectra varies in
shown for example by Beeck et al. (2012) by comparing 3D              one elemental abundance. The χ2 value of the 1D–LTE spectrum
codes. Therefore, it is important that models are probed against      against the 3D models, the NLTE models or observed spectrum
observations. The standard way to test NLTE effects consists          were determined within a spectral range where the 1D–LTE flux
in synthesising individual line profiles and comparing them           changes significantly with the abundance. Finally, the optimal
to very high-resolution spectra (e.g. Allende Prieto et al. 2004;     abundance was determined by minimizing the χ2 value. In ad-
                                                                      dition, the code provides flags to reject any suspicious lines and
 1
   Nearly fully consistent since the computation of 3D atmosphere     upper limit values.
structure assumes LTE.                                                The studied lines were chosen according to two main criteria;
Article number, page 2 of 13
T. Masseron et al.: Probing 3D and NLTE models using APOGEE observations of globular cluster stars

                                                                          2.2. Model spectra
                                                                          For the comparison with the data we used three sets of models:
                                                                          two sets of 1D–NLTE models from two distinct codes and one
                                                                          set of 3D–LTE models from one 3D code.
                                                                               The first set of 1D–NLTE synthetic spectra was derived with
                                                                          the codes TLUSTY and SYNSPEC (Hubeny & Lanz 2017).
                                                                          The atom data and collisional cross-sections are the same
                                                                          as in Osorio et al. (2015) for Mg and in Osorio et al. (2019)
                                                                          for Ca. These atoms have been translated from MULTI to
                                                                          TLUSTY/SYNSPEC format (details in Osorio et al. 2020). The
                                                                          atomic and molecular line lists used will be published in
                                                                          Smith et al. (in prep.), except for Mg where we adopt the data
                                                                          from the model atom reference. For the Ca lines we adopted the
                                                                          parameters from Yu & Derevianko (2018) for the model atom
                                                                          and the atomic line list. The stellar parameters and abundances
                                                                          used are described in Sect. 2.1.
                                                                               The second set of NLTE synthetic spectra was computed
Fig. 1: Illustration of an observed Mg line (upper pan-                   with the online synthesis tool from Kovalev et al. (2018). The
els) and an Fe line (lower panels) and their respective 1D–               radiative transfer code of that tool uses the model atoms and col-
LTE (red) and 1D–NLTE (blue) models in the solar spec-                    lisional cross-sections described in Bergemann et al. (2015) for
trum (Livingston & Wallace 1991). The left panels show high-              Mg and in Bergemann et al. (2013) for Si. For the stellar pa-
resolution (R> 100000) and very high signal-to-noise ratio                rameters the online tool could not provide as many spectra as
(S/N>800) observations, while the right panels show the same              we require in a reasonable time. Therefore, we chose to com-
spectrum at lower resolution (R≈ 22000) and lower signal-to-              pute only three representative effective temperatures for each
noise ratio (S/N = 100). Although the high-resolution spectra             cluster metallicity, such that Teff = 4100K, 4500K, and 4700K.
clearly highlight some modelling issues around the core of the            Corresponding values of log g and microturbulence velocities
lines, the effect is less clearly visible in the lower resolution spec-   have been interpolated from the observed data of the clusters
trum.                                                                     (Sec. 2.1).
                                                                          It should be noted that both 1D–NLTE sets use the same model
                                                                          atmospheres (MARCS Gustafsson et al. 2008), but with differ-
                                                                          ences in the line lists. Nevertheless, we can assume (for close
                                                                          enough log gf values as is true for all the lines in this work) that
they need to be strong enough to appear (and thus be measur-              the differences in the line lists cancel out when the corrections
able) in our sample and they need to have the most reliable loggf         are computed consistently with the same code and the same line
and collisional broadening parameters so that the most accurate           lists. In contrast, the two sets of 1D–NLTE models have notice-
abundance is obtained for the observations.                               able differences regarding the implementation of the NLTE cor-
                                                                          rection. The NLTE calculations performed for this work is the
                                                                          first application of NLTE radiative transfer calculations for sev-
                                                                          eral atomic species simultaneously in cool stars (Osorio et al.
2.1. Observational data
                                                                          2020). The calculations from Kovalev et al. (2018) are tradi-
                                                                          tional NLTE calculations (i.e. one atomic species is calculated
The observational data are from the SDSS IV/APOGEE2 survey                at a time). The two calculations also differ in the electron and
(Gunn et al. 2006; Blanton et al. 2017; Majewski et al. 2017).            hydrogen collisional data, in particular for the levels involved in
The data were reduced as described in Nidever et al. (2015). The          the transitions in the H band, which directly affect the statisti-
selected sample corresponds to GCs members as selected by                 cal balance of those levels. While for electron collisional exci-
Masseron et al. (2019) and Mészáros et al. (2020), with a min-            tation we used the CCC calculations from Barklem et al. (2017)
imum signal-to-noise ratio of 50 and with effective temperatures          for Mg I, Bergemann et al. used the formula from Seaton (1962)
between 4000 K and 4700 K. The final observational sample                 taken from Zhao et al. (1998). For hydrogen collisional excita-
comprises a total of 673 spectra of red giants across 25 glob-            tion between the low-lying levels of Mg I, the same data were
ular clusters.                                                            used in Bergemann et al. and our works (Barklem et al. 2012);
The effective temperatures, surface gravities, microturbulence            instead, for the remaining transitions we used the free electron
velocities, and abundances were adopted from Masseron et al.              model from Kaulakys (1991), while in Bergemann et al. no data
(2019) and Mészáros et al. (2020). However, both works noted              are reported.
some systematics in their Fe abundance scale compared to the                   Zhang et al. (2016) and Zhang et al. (2017) also made NLTE
literature values. Therefore, we adopt the metallicities derived          predictions for Si and Mg lines in the H band, but they did not
by Carretta et al. (2009a,b), except for the ω Cen stars, which           provide the synthesis that would permit a direct comparison such
have been disregarded. We recall that the effective temperatures          as the one we discuss below. Nevertheless, their Mg model atom
were determined from photometry instead of the APOGEE data                and collisional data were identical to the Kovalev values; their
releases values, thus independently of the observed spectra to re-        model atmospheres and statistical equilibrium code were also
duce as much as possible covariances between metallicities and            the same. Therefore, we expect nearly identical results to the
temperatures. Surface gravities were determined via isochrones,           Kovalev results presented in this work. Similarly, Zhang et al.
which have been demonstrated to be more robust against NLTE               (2016) use comparable procedures and data to those used in
effects (Mucciarelli et al. 2015).                                        Kovalev et al. (2018), although they adopt a different scaling fac-
                                                                                                                  Article number, page 3 of 13
A&A proofs: manuscript no. 39484corr

Fig. 2: Synthesis of the Mg lines used in this work. In each sub-panel the upper part shows the 1D–LTE synthesis (dashed black line),
the 1D–NLTE synthesis from this work (red) and Kovalev et al. (2018) (blue), and the 3D–LTE synthesis (green) for a Teff=4500K,
logg=2.5, [M/H]=−1.0 model. The lower parts display the corresponding differences in abundances of the same line between the
1D–LTE model and the NLTE or 3D models for various temperatures and metallicities (large symbols Teff=4000K, logg=1.5;
medium-size symbols Teff=4500K, logg=2.5; small symbols Teff=5000K, logg=2.5).

tor Sh =0.01 in the Drawin formula for H collisions while Ko-       lines we are discussing here have small predicted NLTE effects,
valev adopts 1. Nevertheless, as we show in the following, the Si   leading to negligible differences between those two works.

Article number, page 4 of 13
T. Masseron et al.: Probing 3D and NLTE models using APOGEE observations of globular cluster stars

Fig. 3: Synthesis of Ca lines used in this work. In each sub-panel the upper part shows the 1D–LTE synthesis (dashed black line),
the 1D–NLTE synthesis from this work (red), and the 3D–LTE synthesis (green) for a Teff=4500K, logg=2.5, [M/H]=−1.0 model.
The lower parts display the corresponding differences in abundances of the same line between the 1D–LTE model and the NLTE
or 3D models for various temperatures and metallicities (large symbols Teff=4000K, logg=1.5; medium symbols Teff=4500K,
logg=2.5; small symbols Teff=5000K, logg=2.5).

    The set of 3D spectral syntheses has been taken                     The figures also display in the lower panels the expected offset
from Ludwig et al. (in prep.) based on the 3D radiative-                in abundance between the 3D or NLTE models and the 1D–LTE
hydrodynamical atmosphere grid of Ludwig et al. (2009). Un-             synthesis for different metallicities (−2.0, −1.0, 0.0) and differ-
fortunately, there are only nine models corresponding to gi-            ent {Teff , log g} pairs ({5000,2.5}, {4500,2.5}, {4000,2.5}). The
ants ({Teff , log g} = {5000, 2.5}, {4500, 2.5}, {4000, 1.0} for four   abundance offsets in these figures is equivalent to ∆ log(A) =
metallicities [M/H] = −3.0, −2.0, −1, 0, 0.0).                          log(A)1D/LTE − log(A)NLTE,3D. We also note that the models were
Finally, we note that although the line lists between the 3D and        also degraded to the APOGEE resolution (i.e. R ≈ 22500) to
NLTE computations may not always be rigorously identical, the           ensure that the χ2 minimisation process was homogeneous with
1D–LTE mini-grids use the same corresponding line lists so that         the abundance analysis of the observations.
the computed abundance corrections are fully consistent.

                                                                        2.3.1. Mg
2.3. Comparison between theoretical models
                                                                        From Fig. 2 we can observe two main expected behaviours re-
Figures 2, 3, 4, and 5 illustrate the model outputs respectively for    garding the selected Mg lines. The first group (encompassing the
the selected Mg, Ca, Si, and Fe lines in the H band. All of the fig-    15954 and 16365Å lines) displays very small NLTE effects (for
ures compare the 3D or NLTE synthetic line profiles against 1D–         our computations and for Kovalev’s), but significant 3D effects
LTE synthesis for a Teff=4500K, logg=2.5, [M/H]=−1.0 model,             strongly depending on metallicity. On the contrary, the second
typical of our sample stars. In these figures we note that all 1D       group of lines (15740, 15749, 15765Å) has large negative 3D
spectra have been convolved by a macroturbulence of 3km/s with          effects but larger NLTE effects which largely enhance the line
a rad-tan kernel to match the 3D simulation velocity field.             core intensity, thus leading to an overestimation of a 1D–LTE
                                                                                                               Article number, page 5 of 13
A&A proofs: manuscript no. 39484corr

Fig. 4: Synthesis of Si lines used in this work. In each sub-panel the upper part shows the 1D–LTE synthesis (dashed black
line), the 1D–NLTE synthesis from Kovalev et al. (2018) (blue), and the 3D–LTE synthesis (green) for a Teff=4500K, logg=2.5,
[M/H]=−1.0 model. The lower parts display the corresponding differences in abundances of the same line between the 1D–LTE
model and the NLTE or 3D models for various temperatures and metallicities (large symbols Teff=4000K, logg=1.5; medium-size
symbols Teff=4500K, logg=2.5; small symbols Teff=5000K, logg=2.5).

analysis. The similarities in the second group is not surprising as    2.3.3. Si
the three lines belong to the same triplet. Therefore, their NLTE
effects are expected to be very similar, while their relatively        The selected Si lines show moderate NLTE and large negative
similar strength make them equally sensitive to 3D effects.            3D effects (Fig. 4). Actually, the 15376Å line does not show any
Furthermore, we can use Mg lines to compare NLTE predic-               NLTE effect, probably due to the fact that it is a forbidden tran-
tions for two available codes and methodology. The signs of            sition. The 15960, 16060, and 16094Å lines and the 16215 and
corrections generally agree between the two codes, but the             16828Å are belonging to same triplets. Moreover, their respec-
amplitude may vary. The 15954Å line and to a lesser extent the         tive strengths are similar, naturally leading to quite similar be-
16365Å line are of particular interest because the Kovalev et al.      haviour of the abundances offsets.
(2018) models predict significant corrections while our models
do not. We strongly suspect that the discrepancies are due to          2.3.4. Fe
the difference in the collisional data prescriptions. We also
note that those of the 15954.5Å line show a small blend of a           According to Kovalev’s models, the NLTE corrections for the
Ti I line on its left wing. In addition to being weak, we note         Fe lines in H band are small (Fig. 5). Despite the large number
that this line is included in all our line-by-line syntheses, thus     of lines under study, it appears that the 3D corrections all show
taken consistently into account when applying our differential         negative offsets independently of temperature and metallicity,
approach.                                                              thus suggesting that the 3D effects lead to an overestimation of
                                                                       the mean Fe abundance.

                                                                           After examining the expected behaviours of the lines relative
                                                                       to the theoretical predictions of 3D and NLTE models, we can
                                                                       now use this information to isolate lines with a dominant NLTE
2.3.2. Ca                                                              or 3D effect.

                                                                       3. Discussion
In contrast to Mg, none of the four Ca lines shows strong NLTE         After reviewing the 3D and NLTE model expectations for indi-
effects (Fig. 3), well in line with those reported by Osorio et al.    vidual lines in Sect. 2.2, we now discuss them against the data.
(2020). However, they display large 3D effects, only slightly cor-     Although the resolution of the data does not allow a detailed line
related with metallicity. Moreover, the sign and amplitude of the      profile comparison, the effects still bias the abundance determi-
abundance bias is nearly identical for all lines. This is easily ex-   nation. In this section we use two ways to test the model predic-
plained because three of them belong to the same triplet transi-       tions. The first compares the abundances obtained between two
tion and the last one is the corresponding singlet transition.         different lines. This method provides only relative estimates of
Article number, page 6 of 13
T. Masseron et al.: Probing 3D and NLTE models using APOGEE observations of globular cluster stars

Fig. 5: Synthesis of Fe lines used in this work. In each sub-panel the upper part shows the 1D–LTE synthesis (dashed black line), the
3D–LTE synthesis (green), and for some sub-panels the 1D–NLTE synthesis from Kovalev et al. (2018) (blue) for a Teff=4500K,
logg=2.5, [M/H]=−1.0 model. The lower parts display the corresponding differences in abundances of the same line between the
1D–LTE model and the 3D model for various temperatures and metallicities (large symbols Teff=4000K, logg=1.5; medium-size
symbols Teff=4500K, logg=2.5; small symbols Teff=5000K, logg=2.5).

the 3D and NLTE effects, and we have applied it to the Mg and       3.1. Probing relative 3D and NLTE effects
Si lines in our GCs stars sample. The second way consists in as-
suming that the absolute abundance of each star is known. In this   In Fig. 6 we compare the difference in abundances between two
regard, GCs stars offer an adequate sample as they are known to     Mg lines (Sect. 2.3.1) in the data and for the 3D–LTE mod-
have generally homogeneous abundances of Ca and Fe, unlike          els and the 1D–NLTE models. In this comparison we use the
Mg and Si. This method provides absolute estimates of the 3D        15740Å line as a common reference line between all the pan-
and NLTE effects.                                                   els. In Sect. 2.3.1 we explain that similar 3D effects and similar
                                                                    NLTE effects are also expected between the reference line and
                                                                    the 15749 and 15765Å lines. Although the quantum properties
                                                                                                            Article number, page 7 of 13
A&A proofs: manuscript no. 39484corr

Fig. 6: Differences in abundance obtained between two Mg lines. The black error bars show the star-to-star median of the 1D–LTE
analysis of the APOGEE data. The red squares in the left panels show the prediction obtained with the 1D–NLTE models from this
work. The blue squares in the middle panels show the prediction obtained with the 1D–NLTE models of Kovalev et al. (2018). The
green circles in the right panels show the prediction obtained by the 3D–LTE models of Ludwig et al. (in prep.).

of those lines are identical, we observe in Fig. 6 that the 3D and      It is also interesting to note that in all panels of this figure that the
NLTE models still predict some residual behaviour probably due          predicted dispersion around the trends is relatively small even
to the slight difference in their strengths, suggesting slightly dif-   though all the stars have various stellar parameters and Mg abun-
ferent formation depths in the atmospheres. Both NLTE models            dances, and that we consistently took this into account in our cal-
predict a decreasing trend with increasing metallicity, which ap-       culations. Therefore, we conclude that the dependence of NLTE
pears compatible with the data within the error bars. However,          effect for those Mg lines is dominated by metallicity.
the 3D–LTE models predict a trend in the opposite direction
which does not seem compatible with the data.                               Regarding the Si lines, we observe a similar situation to that
                                                                        for the Mg lines (Fig. 7). There are three sets of lines: two
Concerning the two other Mg lines (15954 and 16364.8Å) the
interpretation of their abundance against the reference line is         sets of triplet transitions (15960, 16060, 16094 Å) and (16215,
more complex because these lines are expected to be 3D sensi-           16828Å), and a forbidden transition (15376 Å). We chose again
tive but not NLTE sensitive, whereas the reference line is sen-         one of the triplet lines as a reference line (15960Å). As for
sitive to both. The data of 15954−15740 differences and the             the Mg triplet lines, the observed 16060−15960 and the 16094-
16365−15740 differences do not show any offset, but do show             15960 differential abundances are nearly null with a slight resid-
a weak positive trend that is reproduced by our 1D–NLTE pre-            ual trend. The predicted NLTE corrections seem compatible with
dictions but not by the Kovalev predictions. This implies either        the data. The agreement between the data and the NLTE models
that Kovalev’s NLTE effects in a 3D hydrodynamical atmosphere           appears particularly good for the 16094−15960, and a bit less
environment are drastically different, or that the NLTE effects of      in the 16060−15960 case. The differential abundance against the
this work cancel out between the two lines, but that the 3D ef-         other lines of the other triplet (16215, 16828Å) not surprisingly
fects are smaller than predicted. In this case, only self-consistent    gives very similar results between the two lines, and is still in
3D–NLTE simulations could sort out the correct interpretation.          very close agreement with Kovalev’s NLTE simulations.
                                                                        The Kovalev predictions are in good agreement with the Si lines
Article number, page 8 of 13
T. Masseron et al.: Probing 3D and NLTE models using APOGEE observations of globular cluster stars

Fig. 7: Differences in abundance obtained between two Si lines. The black error bars show the star-to-star median of the 1D–LTE
analysis of the APOGEE data. The blue squares in the left panels show the prediction obtained with the 1D–NLTE models of
Kovalev et al. (2018). The green circles in the right panels show the prediction obtained by the 3D–LTE models of Ludwig et al.
(in prep.).

Fig. 8: Upper panels: Abundances obtained for individual Ca lines. The black error bars show the star-to-star median of the 1D–LTE
analysis of the APOGEE data. The grey open circles and grey dots are literature data from respectively Roederer et al. (2014) and
Bensby, T. et al. (2014) from optical dwarf spectra. Bottom panels: Corrections obtained with the 1D–NLTE models from this work
(red squares) with the 3D–LTE models of Ludwig et al. (in prep.) (green circles).

abundance data while they are not in good agreement with the        were neglected; for Si, the Kovalev et al. (2018) code uses the
Mg lines abundance data. Beyond the fact that they are two dis-     model atom of Bergemann et al. (2013). This latter work used
tinct atoms, this could also be explained by differences in the     one scaling relation of the Drawin formula from Lambert (1993)
quality and completeness of the collisional data adopted: for Mg,   to assign cross-section values for all energy levels and check it
the Kovalev et al. (2018) code uses the Bergemann et al. (2015)     against some lines of the solar spectrum. Given that the test lines
model where hydrogen collisions between high excitation levels      of Bergemann et al. (2013) and our lines have similar energy lev-

                                                                                                            Article number, page 9 of 13
A&A proofs: manuscript no. 39484corr

Fig. 9: Difference of abundance of Fe lines obtained from individual lines with the literature metallicity of the cluster (black error
bars). The green circles show the prediction of the 3D–LTE models correction and blue squares indicate 1D–NLTE corrections from
Kovalev et al. (2018).

els, this may explain why Kovalev’s NLTE predictions provide               Regarding the 3D predictions, they show moderate ampli-
good agreement with the data. However, our study also involves         tude with a negative dependence with metallicity, a trend which
the forbidden transition at 15376 Å. For that line, the data seem      seems hardly compatible with the data. The case of the 15376Å
to disagree with Kovalev’s predictions, suggesting that the scal-      line is of particular interest because it is a forbidden transition
ing relation they used to compute the Si cross-sections is not         whose NLTE effects can be considered negligible. Therefore, the
valid for all transitions. This supports the idea that the scaling     NLTE effect should only be due to the 16960Å line alone. The
relation in NLTE calculations should be avoided and, as quoted         data suggest a v-shaped dependence against metallicity, and so
from Bergemann et al. (2013), that ‘more accurate estimates of         do the NLTE simulations. However, there is an offset between
the collision rates are desirable and, once available, will signifi-   the models and the data that might be explained by a combi-
cantly improve the accuracy of the calculations’.                      nation of the NLTE effects with the 3D effects. Therefore, the
                                                                       3D–LTE spectra for the weak lines do not provide a good pre-

Article number, page 10 of 13
T. Masseron et al.: Probing 3D and NLTE models using APOGEE observations of globular cluster stars

diction of reality, and 3D–NLTE should be more appropriate.            estimates because the giants used in the Amarsi and Monshon-
To verify this hypothesis a full 3D–NLTE calculation would be          kina works are not identical to our GCs stars, and it is not clear
necessary. We finally note that the microturbulence and macro-         whether the Fe lines between the studies are fully compatible. In
turbulence velocities in the 1D–LTE calculations could be tuned        addition, as we demonstrate in the case of Mg and Si in Sect. 3.1,
to match the 3D computations. However, while in principle these        all these results are strongly dependent on the current knowledge
empirical parameters can be freely varied, the values we use in        of the Fe atom and its cross-section values, which are known to
this work are compatible with optical studies and with APOGEE          be still quite poor.
spectra standard analysis (Holtzman et al. 2015). Moreover, we              Moreover, the unrealistically large predicted 3D correc-
argue that if this value had to be changed, it would have a drastic    tions also highlight the problem of applying corrections self-
impact on other strong lines such as the one in Figs. 2 and 4.         consistently. An illustration of this argument can be seen when
                                                                       looking at the Fe lines in Fig. 9, which show the difference be-
                                                                       tween the Fe abundances and the literature cluster metallicity
3.2. Probing absolute 3D/NLTE effects                                  (Carretta et al. 2009a,b). In this figure the 3D–LTE estimated
Figure 8 shows the line-by-line abundances of Ca obtained for          corrections are overplotted and predict a negative offset, which
the whole sample and is compared to the analysis of opti-              is the opposite of the data. If we had to derive the metallic-
cal dwarf spectra of Roederer et al. (2014) and Bensby, T. et al.      ity of the 3D models with 1D models, we would certainly find
(2014). All data sets show a global decreasing trend with in-          a lower metallicity. By extension, we can also derive differ-
creasing metallicity as expected by chemical evolution. The 3D         ent Teff and log g (see also Chiavassa et al. 2011, where different
and NLTE expected corrections are also displayed. As explained         1D effective temperatures are derived in 3D supergiant model
in Sect. 2.3.2, the models predictions behave very similarly be-       spectra depending on whether you use optical molecular band
tween the three Ca lines, with nearly null NLTE corrections            strength or infrared flux). In other words, the derived stellar pa-
and large 3D corrections. The agreement between the optical            rameters derived with 3D models may be different to the one
data and the IR data in Fig. 8 suggest that the 3D corrections         derived in 1D, implying that 3D abundance corrections applied
should not be larger than 0.1 dex overall. Another possibility         on 1D abundances with the same stellar parameters between 1D
would imply that optical data are affected by NLTE while IR            and 3D are not consistent; thus, a direct comparison (as in Fig. 8
data are affected by 3D of the same amplitude. However, 3D             and Fig. 9) may not be adequate. Actually, if we had to apply
abundance corrections are known to be larger in dwarfs than            the average ∼-0.3 metallicity correction from Fig. 9 in addition
in giants (Ludwig et al. 2008) in contradiction with the fact that     to the ∼-0.3 Ca corrections in Fig. 8, the net resulting correc-
the optical data are composed of dwarfs while the IR are com-          tions would be at more realistic levels. Nevertheless, we note
posed of giants. Moreover, we argue that if the 3D corrections         that self-consistency is certainly an important argument for our
as estimated in this work were to be applied (and so for NLTE          discussion for probing absolute corrections as we do for Ca and
in the optical data), the chemical evolution of Ca would be af-        Fe, but that does not affect our method for testing relative correc-
fected in an unrealistic way that is incompatible with chem-           tions (Sect. 3.1) where the line-differential approach ensures that
ical evolution expectations. The line-differential work we did         the differences in stellar parameters between 1D and 3D cancel
for Mg and Si (see Sect. 3.1) certainly supports to some ex-           out.
tent that 3D effects on abundances are not accurate enough, or at
least require that NLTE effects be simultaneously taken into ac-       4. Conclusions
count. This conclusion also corroborates theoretical predictions
where 3D–LTE abundances are often worse than 1D–LTE values             In this paper we have tested the theoretical predictions of some
(Nordlander, T. et al. 2017) and is related to the NLTE-masking        3D–LTE and 1D–NLTE models against a homogeneous analysis
effect described by (Rutten & Kostik 1982).                            of a large sample of GC spectra from the APOGEE large spec-
     Regarding Fe, it is striking in Fig. 9 that the data system-      troscopic survey. In particular, we have discussed the expected
atically show a positive offset of ∼0.1 dex, implying that the         effects on the individual lines abundances of four elements (Mg,
overall Fe abundance is overestimated compared to the literature.      Si, Ca, and Fe). Using Mg lines, we could compare two distinct
First of all, this bias in metallicity is not likely due to our spe-   sets of models for 1D–NLTE and show that the predictions are in
cific procedure because a similar metallicity offset is observed       the same direction. Nevertheless, we observe some quantitative
in other works with distinct temperature scales dealing with the       differences that are mainly due to the inclusion or not of good
APOGEE spectra (Mészáros et al. 2013; Holtzman et al. 2015;            and complete collisional cross-sections, and not by the differ-
Mészáros et al. 2015; Masseron et al. 2019). Figure 9 shows that       ences in the codes themselves.
NLTE effects do not seem to significantly affect the Fe lines          The analysis of the abundance differences between several Mg
in the H band. The literature reports that optical Fe lines also       lines (and similarly Si lines) allows us to also verify the impact
do not seem to be affected by NLTE. Carretta et al. (2009a,b)          of the 3D effects. However, some 3D–LTE predictions seem to
checked that their metallicity determined with optical Fe I and        provide a satisfactory explanation for some of the lines abun-
with Fe II lines are in agreement, suggesting that NLTE effects        dance behaviour, but most do not. Nevertheless, while waiting
are negligible in the optical. In contrast, recent self-consistent     for more extensive 3D model grids, our work supports and gen-
3D–NLTE models of one very metal-poor giant (Amarsi et al.             eralises the fact that 1D–LTE provides a better approximation
2016) estimate that metallicities determined from optical Fe I         than 3D–LTE, and thus that 1D–LTE calculation partially cancel
and Fe II lines are respectively underestimated by 0.17 and 0.08       out the 3D–NLTE effects (Asplund et al. 2003). We reach simi-
dex. In a similar way, Mashonkina et al. (2016b) compute 1D–           lar conclusions regarding Ca abundances where the 3D–LTE es-
NLTE corrections for many optical Fe lines in giants and pro-          timates on the Ca abundances seem too large. However, we rec-
vide corrections of ∼0.1 dex at low metallicities. Such correc-        ommend strong caution when applying 3D corrections directly
tions onto the GCs metallicities of Carretta’s work may suggest        to 1D abundances without assuming self-consistent 3D and 1D
that the APOGEE data provide quite accurate GC metallicities.          stellar parameters.
But before drawing any firm conclusion, these are quite coarse         Finally, we have checked the Fe abundances for 12 Fe lines and
                                                                                                              Article number, page 11 of 13
A&A proofs: manuscript no. 39484corr

show a systematic offset against the literature metallicity values.                  Amarsi, A. M., Lind, K., Asplund, M., Barklem, P. S., & Collet, R. 2016, MN-
Although more NLTE models with accurate atom data for Fe                                RAS, 463, 1518
                                                                                     Asplund, M., Carlsson, M., & Botnen, A. V. 2003, A&A, 399, L31
lines are crucially needed, our results may suggest that the glob-                   Asplund, M., Nordlund, Å., Trampedach, R., Allende Prieto, C., & Stein, R. F.
ular clusters metallicity maybe underestimated in the optical-                          2000a, A&A, 359, 729
based literature, whereas the Fe lines of the H band offer more                      Asplund, M., Nordlund, Å., Trampedach, R., & Stein, R. F. 2000b, A&A, 359,
accurate values.                                                                        743
                                                                                     Barklem, P. S., Belyaev, A. K., Spielfiedel, A., Guitou, M., & Feautrier, N. 2012,
In the end our study suggests that the largest effect on abun-                          A&A, 541, A80
dances for late-type giants is NLTE, and that 3D is only a second-                   Barklem, P. S., Osorio, Y., Fursa, D. V., et al. 2017, A&A, 606, A11
order effect that cannot be treated separately. This idea finds the-                 Beeck, B., Collet, R., Steffen, M., et al. 2012, A&A, 539, A121
oretical support in the detailed study Amarsi et al. (2016) where                    Bensby, T., Feltzing, S., & Oey, M. S. 2014, A&A, 562, A71
                                                                                     Bergemann, M. 2014, Analysis of Stellar Spectra with 3-D and NLTE Models,
1D–LTE, 1D–NLTE, 3D–LTE, and 3D–NLTE abundances of Fe                                   187–205
lines in a few giants are compared.                                                  Bergemann, M., Kudritzki, R.-P., Gazak, Z., Davies, B., & Plez, B. 2015, ApJ,
    The line-differential technique we develop here to probe                            804, 113
NLTE and/or 3D models can be applied to any other elements                           Bergemann, M., Kudritzki, R.-P., Plez, B., et al. 2012a, ApJ, 751, 156
                                                                                     Bergemann, M., Kudritzki, R.-P., Würl, M., et al. 2013, ApJ, 764, 115
and other data sets, and offers the great advantage of being free                    Bergemann, M., Lind, K., Collet, R., Magic, Z., & Asplund, M. 2012b, MNRAS,
from knowing the abundance a priori. Nevertheless, to be valid                          427, 27
this method still requires three things, that the stellar parameters                 Bergemann, M. & Nordlander, T. 2014, NLTE Radiative Transfer in Cool Stars,
are well constrained, that the line’s oscillator strength and colli-                    169–185
                                                                                     Blanton, M. R., Bershady, M. A., Abolfathi, B., et al. 2017, AJ, 154, 28
sional broadening coefficient are accurately known, and that the                     Caffau, E., Bonifacio, P., Oliva, E., et al. 2019, A&A, 622, A68
selected lines are relatively clean of blends.                                       Caffau, E., Ludwig, H. G., Steffen, M., et al. 2008, A&A, 488, 1031
                                                                                     Carretta, E., Bragaglia, A., Gratton, R., & Lucatello, S. 2009a, A&A, 505, 139
Acknowledgements. We deeply thank H.G. Ludwig for his careful checks of the
                                                                                     Carretta, E., Bragaglia, A., Gratton, R. G., et al. 2009b, A&A, 505, 117
3D spectra and thoughtful comments. We thank M. Kovalev and M. Bergemann
                                                                                     Chiavassa, A., Freytag, B., Masseron, T., & Plez, B. 2011, A&A, 535, A22
for their great support of their NLTE online synthesis tool. We also thank A.
                                                                                     Collet, R., Asplund, M., & Trampedach, R. 2006, ApJ, 644, L121
Amarsi for fruitful discussion about 3D/NLTE effects. T.M. acknowledges sup-         De Silva, G. M., Freeman, K. C., Bland-Hawthorn, J., et al. 2015, MNRAS, 449,
port from Spanish Ministry of Economy and Competitiveness (MINECO) under
                                                                                        2604
the 2015 Severo Ochoa Program SEV-2015-0548. T.M., D.A.G.H. and O.Z. also            Ezzeddine, R., Merle, T., Plez, B., et al. 2018, A&A, 618, A141
acknowledge support from the State Research Agency (AEI) of the Spanish Min-         Freytag, B., Steffen, M., Ludwig, H. G., et al. 2012, Journal of Computational
istry of Science, Innovation and Universities (MCIU) and the European Regional          Physics, 231, 919
Development Fund (FEDER) under grant AYA2017-88254-P. Y.O, and C.A.P.                Galsgaard, K. & Nordlund, Å. 1996, J. Geophys. Res., 101, 13445
acknowledge support from the same agency under grant AYA2017-86389-P.                Gilmore, G., Randich, S., Asplund, M., et al. 2012, The Messenger, 147, 25
SzM has been supported by the János Bolyai Research Scholarship of the Hun-          Gunn, J. E., Siegmund, W. A., Mannery, E. J., et al. 2006, AJ, 131, 2332
garian Academy of Sciences, by the Hungarian NKFI Grants K-119517 and                Gustafsson, B., Edvardsson, B., Eriksson, K., et al. 2008, A&A, 486, 951
GINOP-2.3.2-15-2016-00003 of the Hungarian National Research, Development            Hawkins, K., Masseron, T., Jofré, P., et al. 2016, A&A, 594, A43
and Innovation Office, and by the ÚNKP-19-4 New National Excellence Pro-             Holtzman, J. A., Shetrone, M., Johnson, J. A., et al. 2015, AJ, 150, 148
gram of the Ministry for Innovation and Technology. This paper made use of the       Hubeny, I. & Lanz, T. 2017, arXiv e-prints, arXiv:1706.01859
IAC Supercomputing facility HTCondor (http://research.cs.wisc.edu/htcondor/),        Kaulakys, B. 1984, Journal of Physics B: Atomic, 17, 4485
partly financed by the Ministry of Economy and Competitiveness with FEDER            Kaulakys, B. 1991, Journal of Physics B Atomic Molecular Physics, 24, L127
funds, code IACA13-3E-2493. The authors also acknowledge the technical ex-           Klevas, J., Kučinskas, A., Steffen, M., Caffau, E., & Ludwig, H. G. 2016, A&A,
pertise and assistance provided by the Spanish Supercomputing Network (Red              586, A156
Española de Supercomputación), as well as the computer resources used: the La-       Korn, A. J., Shi, J., & Gehren, T. 2003, A&A, 407, 691
Palma Supercomputer, located at the Instituto de Astrofísica de Canarias. Fund-      Korotin, S. A., Andrievsky, S. M., Caffau, E., Bonifacio, P., & Oliva, E. 2020,
ing for the Sloan Digital Sky Survey IV has been provided by the Alfred P. Sloan        MNRAS, 496, 2462
Foundation, the U.S. Department of Energy Office of Science, and the Partici-        Kovalev, M., Bergemann, M., Ting, Y.-S., & Rix, H.-W. 2019, A&A, 628, A54
pating Institutions. SDSS acknowledges support and resources from the Center         Kovalev, M., M. Bergemann, Brinkmann, S., & and MPIA IT-department. 2018,
for High-Performance Computing at the University of Utah. The SDSS web site             NLTE MPIA web server, [Online]. Available: http://nlte.mpia.de Max Planck
is www.sdss.org. SDSS is managed by the Astrophysical Research Consortium               Institute for Astronomy, Heidelberg.
for the Participating Institutions of the SDSS Collaboration including the Brazil-   Lambert, D. L. 1993, Physica Scripta, T47, 186
ian Participation Group, the Carnegie Institution for Science, Carnegie Mellon       Lind, K., Amarsi, A. M., Asplund, M., et al. 2017, MNRAS, 468, 4311
University, the Chilean Participation Group, the French Participation Group,         Lind, K., Melendez, J., Asplund, M., Collet, R., & Magic, Z. 2013, A&A, 554,
Harvard-Smithsonian Center for Astrophysics, Instituto de Astrofísica de Ca-            A96
narias, The Johns Hopkins University, Kavli Institute for the Physics and Math-      Livingston, W. & Wallace, L. 1991, An atlas of the solar spectrum in the infrared
ematics of the Universe (IPMU) / University of Tokyo, the Korean Participation          from 1850 to 9000 cm-1 (1.1 to 5.4 micrometer)
Group, Lawrence Berkeley National Laboratory, Leibniz Institut für Astrophysik       Ludwig, H. G., Caffau, E., Steffen, M., et al. 2009, Mem. Soc. Astron. Italiana,
Potsdam (AIP), Max-Planck-Institut für Astronomie (MPIA Heidelberg), Max-               80, 711
Planck-Institut für Astrophysik (MPA Garching), Max-Planck-Institut für Ex-          Ludwig, H.-G., González Hernández, J. I., Behara, N., Caffau, E., & Steffen,
traterrestrische Physik (MPE), National Astronomical Observatories of China,            M. 2008, in American Institute of Physics Conference Series, Vol. 990, First
New Mexico State University, New York University, University of Notre Dame,             Stars III, ed. B. W. O’Shea & A. Heger, 268–272
Observatório Nacional / MCTI, The Ohio State University, Pennsylvania State          Ludwig, H.-G., Koesterke, L., & Allende Prieto, C. in prep.
University, Shanghai Astronomical Observatory, United Kingdom Participation          Magic, Z., Collet, R., Asplund, M., et al. 2013, A&A, 557, A26
Group, Universidad Nacional Autónoma de México, University of Arizona, Uni-          Majewski, S. R., Schiavon, R. P., Frinchaboy, P. M., et al. 2017, AJ, 154, 94
versity of Colorado Boulder, University of Oxford, University of Portsmouth,         Mashonkina, L. 2014, in IAU Symposium, Vol. 298, Setting the scene for Gaia
University of Utah, University of Virginia, University of Washington, University        and LAMOST, ed. S. Feltzing, G. Zhao, N. A. Walton, & P. Whitelock, 355–
of Wisconsin, Vanderbilt University, and Yale University.                               365
                                                                                     Mashonkina, L., Gehren, T., Shi, J. R., Korn, A. J., & Grupp, F. 2011, A&A, 528,
                                                                                        A87
                                                                                     Mashonkina, L., Ludwig, H. G., Korn, A., Sitnova, T., & Caffau, E. 2013, Mem-
References                                                                              orie della Societa Astronomica Italiana Supplementi, 24, 120
                                                                                     Mashonkina, L., Sitnova, T., & Belyaev, A. K. 2017, A&A, 605, A53
Alexeeva, S., Ryabchikova, T., Mashonkina, L., & Hu, S. 2018, ApJ, 866, 153          Mashonkina, L. I., Belyaev, A. K., & Shi, J. R. 2016a, Astronomy Letters, 42,
Allende Prieto, C., Asplund, M., & Fabiani Bendicho, P. 2004, A&A, 423, 1109            366
Allende Prieto, C., Asplund, M., García López, R. J., & Lambert, D. L. 2002,         Mashonkina, L. I., Sitnova, T. N., & Pakhomov, Y. V. 2016b, Astronomy Letters,
   ApJ, 567, 544                                                                        42, 606
Amarsi, A. M., Barklem, P. S., Collet, R., Grevesse, N., & Asplund, M. 2019,         Masseron, T., García-Hernández, D. A., Mészáros, S., et al. 2019, A&A, 622,
   A&A, 624, A111                                                                       A191

Article number, page 12 of 13
T. Masseron et al.: Probing 3D and NLTE models using APOGEE observations of globular cluster stars

Masseron, T., Merle, T., & Hawkins, K. 2016, BACCHUS: Brussels Automatic
   Code for Characterizing High accUracy Spectra, Astrophysics Source Code
   Library
Merle, T., Thévenin, F., Pichon, B., & Bigot, L. 2011, MNRAS, 418, 863
Mészáros, S., Holtzman, J., García Pérez, A. E., et al. 2013, AJ, 146, 133
Mészáros, S., Martell, S. L., Shetrone, M., et al. 2015, AJ, 149, 153
Mészáros, S., Masseron, T., García-Hernández, D. A., et al. 2020, MNRAS, 492,
   1641
Mucciarelli, A., Lapenna, E., Massari, D., et al. 2015, ApJ, 809, 128
Nidever, D. L., Holtzman, J. A., Allende Prieto, C., et al. 2015, AJ, 150, 173
Nissen, P. E. & Gustafsson, B. 2018, A&A Rev., 26, 6
Nordlander, T., Amarsi, A. M., Lind, K., et al. 2017, A&A, 597, A6
Nordlund, Å. & Stein, R. F. 2009, in American Institute of Physics Conference
   Series, ed. I. Hubeny, J. M. Stone, K. MacGregor, & K. Werner, Vol. 1171,
   242–259
Osorio, Y., Allende Prieto, C., Hubeny, I., Mészáros, S., & Shetrone, M. 2020,
   A&A, 637, A80
Osorio, Y. & Barklem, P. S. 2016, A&A, 586, A120
Osorio, Y., Barklem, P. S., Lind, K., et al. 2015, A&A, 579, A53
Osorio, Y., Lind, K., Barklem, P. S., Allende Prieto, C., & Zatsarinny, O. 2019,
   A&A, 623, A103
Pereira, T. M. D., Kiselman, D., & Asplund, M. 2009, A&A, 507, 417
Randich, S., Gilmore, G., & Gaia-ESO Consortium. 2013, The Messenger, 154,
   47
Roederer, I. U., Preston, G. W., Thompson, I. B., et al. 2014, AJ, 147, 136
Rutten, R. J. & Kostik, R. I. 1982, A&A, 115, 104
Sbordone, L., Bonifacio, P., Caffau, E., et al. 2010, A&A, 522, A26
Seaton, M. J. 1962, Proceedings of the Physical Society, 79, 1105
Seaton, M. J. 1962, Atomic and Molecular Processes. Edited by D. R. Bates.
   Library of Congress Catalog Card Number 62-13122. Published by Academic
   Press, 375
Smith, V. V., Bizyaev, D., Allende Prieto, C., Cunha, K., & et .al. in prep.
Spite, M., Andrievsky, S. M., Spite, F., et al. 2012, A&A, 541, A143
Steffen, M., Prakapavičius, D., Caffau, E., et al. 2015, A&A, 583, A57
Černiauskas, A., Kučinskas, A., Klevas, J., et al. 2018, A&A, 615, A173
Vögler, A., Shelyag, S., Schüssler, M., et al. 2005, A&A, 429, 335
Vrinceanu, D. 2005, Phys. Rev. A, 72
Yu, Y. & Derevianko, A. 2018, Atomic Data and Nuclear Data Tables, 119, 263
Zhang, J., Shi, J., Pan, K., Allende Prieto, C., & Liu, C. 2016, ApJ, 833, 137
Zhang, J., Shi, J., Pan, K., Allende Prieto, C., & Liu, C. 2017, ApJ, 835, 90
Zhao, G., Butler, K., & Gehren, T. 1998, A&A, 333, 219

                                                                                                                Article number, page 13 of 13
You can also read